A parameterized physics-informed machine learning approach for solving heat and mass transfer equations in the drying process

Abstract

This paper introduces a novel application for surrogate modeling in the context of coupled heat and mass transfer during drying using physics-informed neural networks (PINNs). The Luikov model is implemented within the PINNs model to simulate heat and mass transfer dynamics. The proposed method enables the integration of crucial parameters like the thermal Biot number (Bit) and mass transfer Biot number (Bim) in the model's architecture, facilitating a versatile framework to address a wide range of operating conditions. The range of variation for the Bit lies between 0 and 1, while the range of the Bim is between 0 and 10. The results predicted by PINNs model has been validated against reference values in literature obtained via Finite Element Method (FEM), analytical solution, and experimental data to verify the model's accuracy. Sensitivity analysis revealed that temperature sensitivity to Bit peaked at an optimal range, while moisture sensitivity remained negligible. The model's capability to predict outcomes across varying Bim values was confirmed, showing high accuracy in capturing the dynamics of the drying process. Sample refinement enhanced prediction accuracy, especially in the critical range of Bim values.